Research Of Motor Bearing Fault Diagnosis Methods Based On Machine Learning

As an important basic component in modern engineering and electromechanical equipment,motor bearing plays a key role in industrial production.With the continuous development towards digitalization and intelligence of industrial electromechanical equipment in our country,it brings more challenges on the ability of motor bearing to ensure the normal operation of electromechanical equipment.If the potential or early failure of motor bearing was not found in time,it will bring great hidden danger to the safety of mechanical work and serious production accident.The thesis aims at motor bearing fault diagnosis,and conducts in-depth analysis and research from three parts:vibration signal preprocessing,feature extraction and fusion,and fault diagnosis model establishment and improvement.In this paper,based on the analyzes of the structure,failure mechanism,frequency characteristics of the bearing,and common time-frequency analysis methods for handling unsteady signals,firstly,in view of the non-linear and non-stationary characteristics of the bearing fault vibration signal,the wavelet transform method was introduced for denoising processing.The two signal decomposition algorithms of empirical mode decomposition and ensemble empirical mode decomposition were compwered and analyzed,and the ensemble empirical mode decomposition with better extraction effect was selected for signal decomposition and reconstruction.The energy entropy of the intrinsic modal components and the fine composite multi-scale dispersion entropy of the reconstructed signal were calculated,and the feature set was constructed by two feature fusion methods of cascade and weighted summation.Secondly,due to the low recognition rate and slow speed of traditional optimization algorithms,the gray wolf optimization(GWO)algorithm was improved by introducing Tent chaotic mapping,nonlinear convergence factors,and introducing disturbance factors from genetic algorithm mutation strategies.Algorithm performance test results show that the improved gray wolf algorithm has a fast convergence speed and a strong ability to jump out of local extremes.Then,the support vector machine(SVM)method was applied to fault diagnosis,and the improved gray wolf algorithm was used to optimize the penalty factor and kernel function parameters of the support vector machine to build a fault diagnosis model with better performance.Afterwards a fault diagnosis model based on the combination of wavelet denoising,feature fusion,improved gray wolf algorithm and support vector machine was established.Finally,based on the above work,the effectiveness of the cascaded fusion feature set,improved gray wolf optimization algorithm and fault diagnosis method used in this paper was verified.The diagnosis models of PSO-SVM,GWO-SVM and improved GWO-SVM is verified through experiments.The comparison results show that the cascade fusion feature set can better characterize the bearing fault characteristics,and the improved gray wolf algorithm has strong optimization capabilities,and the improved GWO-SVM diagnosis model can effectively speed up the diagnosis and improve the accuracy.